SkillCommunity
PyMCNP skillby Fusion Science & Ion Beam Technology Group (LBNL)
A SKILL.md shipped with PyMCNP that teaches a coding agent to write, scan, run and read MCNP inputs and outputs in Python.
Needs a MCNP licence23· updated 3 days ago
MCNP MCP servers and AI agent skills
MCNP has no MCP server, but it has one agent skill: PyMCNP, a Python library from Lawrence Berkeley National Laboratory's Fusion Science and Ion Beam Technology group, ships a SKILL.md that teaches a coding agent to build, scan, run and read MCNP inputs and outputs. You still need your own MCNP installation, which is export-controlled under 10 CFR 810 and distributed only through RSICC.
- No official LANL server yet.
- 1 agent skill listed.
LANL1 skillChecked 4 Nov 2026
Integrations for MCNP
Agent skills
Listings are sorted by origin, then by recent activity. GitHub data is refreshed nightly. Nothing here is ranked.
How AI connects to MCNP
An MCP server or skill can only do what MCNP lets programs do. MCNP exposes 2 automation interfaces:
- MCNP input decks (Input files): An MCNP problem is a text input file of cell, surface and data cards, with materials, sources, tallies and run controls such as NPS or KCODE given as data cards. Lines are limited to 80 columns, which PyMCNP handles by wrapping long cards. LANL publishes the current Theory and User Manual on the MCNP website; the latest release listed there is MCNP6.3.1. Documentation
- PyMCNP (community Python API) (Python): PyMCNP parses MCNP inputs into Python objects and writes them back, generates parameter scans, runs batches of inputs through the mcnp6 executable in parallel, and reads OUTP, MESHTAL and PTRAC output, including streaming very large PTRAC files. It is BSD-licensed (LBNL variant) and supports Python 3.11 to 3.14. It is not part of LANL's distribution. Documentation
What LANL offers
LANL's MCNP website covers releases, manuals, classes and how to obtain the code, and mentions no AI assistant or MCP server as of 4 November 2026. RSICC states that all MCNP versions are controlled under 10 CFR 810, limiting distribution to US persons and citizens of Appendix A countries, with source code needing extra approval. Any AI integration must run where your licensed copy is.
Checked 4 Nov 2026. Sources: MCNP website (LANL), How to get the MCNP code (LANL), RSICC home page and MCNP6.3.1 announcement
What you can do with MCNP and AI
- Build an MCNP input from a description in Python, for example a shielded source with a flux tally, and check that the whole file parsed with nothing left over.
- Scan a geometry or material parameter across many copies of a base input and run them in separate directories with a set number of workers.
- Read OUTP results and report tally means, errors and whether each tally passed MCNP's ten statistical checks before you trust it.
- Stream a large PTRAC file and trace which nuclide produced the photons entering a detector cell.
Taken from the listings' own documentation. Check each listing for exactly what it supports.
Installing
The listings here document setup for Claude Code. Install steps, copied from each project's own documentation, are on each listing's page. Most MCNP integrations run on your own computer next to MCNP, so you need MCNP installed and licensed on the same machine unless the listing says otherwise.
Related tools
Questions
Is there an MCNP MCP server?
No public one was found on 4 November 2026. The PyMCNP skill is the closest: it guides a coding agent that can run Python and call your local mcnp6 executable.
Can I use an AI assistant with MCNP given export controls?
Export control applies to the MCNP code itself, which RSICC distributes under 10 CFR 810 to eligible users. A skill like PyMCNP contains no MCNP code, but where an agent runs and what model service sees your inputs is a question for your export-control officer, not a software setting.
Where do I get MCNP?
From RSICC at Oak Ridge National Laboratory. LANL says the distribution package includes executables for Linux, macOS and Windows; non-US citizens should generally request the executable-only package, and LANL staff use the internal network instead.
Who makes PyMCNP?
The Fusion Science and Ion Beam Technology group at Lawrence Berkeley National Laboratory. It is a separate open-source project, not an official LANL tool.
AI-generated models, drawings and results must be checked by a qualified engineer before use.
Sources (checked 4 Nov 2026): MCNP website (LANL) · How to get the MCNP code (LANL) · MCNP 6.3.1 release documents (LANL) · RSICC home page · FSIBT/PyMCNP README · MCNP product page
MCNP is a trademark of LANL. Tresca is independent and not affiliated with or endorsed by LANL.